[gclist] Re: Advice requested

Hans Boehm boehm@hoh.engr.sgi.com
Wed, 28 Apr 1999 09:03:35 -0700

On Apr 28,  1:05pm, marc.shapiro@inria.fr wrote:
> Distributed GC algorithms for C++ do exist.  Take a look at
> http://www-sor.inria.fr/projects/sspc.  We (INRIA) have an open-source
> implementation that you can use.  However it is tightly integrated into
> a remote-invocation system that is not Corba.  Corba does not support
> GC.  DCOM, does, in a very poor way (explicit reference counting, no
> fault tolerance).
Could you expand on that a little, especially the comment about CORBA?  My
model has always been that, except for promptness issues, a local GC can't
possibly make things worse for these systems.  Since none of them support
anything beyond reference counting, distributed cyclic garbage will leak
anyway.  Given that, it seems like GC finalization should give you a hook
that's equivalent to what you get from explicit deallocation?  But I've never
looked at the details ...


Hans-J. Boehm